BigMoeOnEdge

docs(readme): open on a silent demo loop, with the full video one click away (#297)

#27 -Commit a0bb4a1cda pushed by vrr

main

2026-10-19 07:23:46 +00:00

0s

chore(release): 0.31.0 (#281)

#26 -Commit 4c80400a3d pushed by vrr

v0.31.0

2026-10-18 23:03:29 +00:00

0s

chore(release): 0.31.0 (#281)

#25 -Commit 4c80400a3d pushed by vrr

main

2026-10-19 07:23:46 +00:00

0s

chore(release): 0.30.0 (#259)

#24 -Commit 0d7a8d4d09 pushed by vrr

main

2026-10-18 15:45:10 +00:00

0s

chore(release): 0.29.0 (#228)

#23 -Commit 08c6bfb89c pushed by vrr

v0.29.0

2026-10-17 16:38:24 +00:00

0s

docs(roadmap): Vulkan and OpenCL GPUs, a Connect view for clients, the planner drawn (#258)

#22 -Commit 0207ed9382 pushed by vrr

main

2026-10-17 09:41:39 +00:00

0s

feat(desktop): bmoe-server and a local web UI, for Windows, macOS and Linux (#215)

#21 -Commit 5a07bd6b9a pushed by vrr

main

2026-10-15 03:39:58 +00:00

0s

docs(readme): decisions in a section of their own, routed arena in the NPU prefill (#209)

#20 -Commit 374f562c92 pushed by vrr

v0.28.0

2026-09-30 19:41:05 +00:00

0s

docs(readme): decisions in a section of their own, routed arena in the NPU prefill (#209)

#19 -Commit 374f562c92 pushed by vrr

main

2026-09-30 19:41:04 +00:00

0s

feat: --decide, choose from a list from one prefill, with no decode (#201)

#18 -Commit 1d37057fac pushed by vrr

v0.27.0

2026-09-29 13:36:36 +00:00

0s

feat: prefill on the NPU, decode on the CPU (--prefill-device), in the release APK (#200)

#17 -Commit 10af539a63 pushed by vrr

v0.26.0

2026-09-29 13:36:36 +00:00

0s

feat(moe): Nemotron 3.5 (nemotron_h_moe) and Ornith 1.5 support (#203)

#16 -Commit 50d0a38338 pushed by vrr

v0.25.0

2026-09-29 13:36:35 +00:00

0s

feat: --decide, choose from a list from one prefill, with no decode (#201)

#15 -Commit 1d37057fac pushed by vrr

main

2026-09-29 13:36:31 +00:00

0s

docs(changelog): close 0.24.0

#14 -Commit 74ba18f53d pushed by vrr

v0.24.0

2026-09-09 07:34:17 +00:00

0s

docs(changelog): close 0.24.0

#13 -Commit 74ba18f53d pushed by vrr

main

2026-09-09 07:34:16 +00:00

0s

fix(scripts): bench-report model size and storage probe (du on the raw path) (#194)

#12 -Commit 92a1114917 pushed by vrr

main

2026-09-01 13:34:16 +00:00

0s

docs(assets): label the Qwen3.8-Flash-Next hero clip like the DeepSeek one

#11 -Commit 557c63dda0 pushed by vrr

main

2026-08-31 10:50:32 +00:00

0s

feat(app): Qwen3.8-Flash-Next Q2_K in the catalog, the low-dense build (#189)

#10 -Commit f24bd5aecc pushed by vrr

main

2026-08-31 02:20:01 +00:00

0s

docs(bench): tell contributors to check the mode line before reporting a number (#190)

#9 -Commit 4c02169fee pushed by vrr

main

2026-08-30 13:41:43 +00:00

0s

fix(cli): make the Windows release exe survive a double click (#183)

#8 -Commit d754dfd6fc pushed by vrr

v0.23.0

2026-08-30 14:41:40 +00:00

0s

feat(moe): cache-aware expert substitution (--expert-substitute, experimental) and --ppl (#171)

#7 -Commit 4334c89616 pushed by vrr

main

2026-08-29 21:21:44 +00:00

0s

feat(telemetry): add prefill-phase attribution (#175)

#6 -Commit 039b8f1e2f pushed by vrr

main

2026-08-29 13:11:43 +00:00

0s

docs(changelog): Qwen3.8-Flash-Next numbers on the final pin

#5 -Commit 2c73fb3a5d pushed by vrr

v0.22.0

2026-08-29 14:41:44 +00:00

0s

docs(changelog): Qwen3.8-Flash-Next numbers on the final pin

#4 -Commit 2c73fb3a5d pushed by vrr

main

2026-08-28 20:52:47 +00:00

0s

ci: prebuilt bmoe-cli for Linux, macOS and Windows on every release (#178)

#3 -Commit f9371b4ddb pushed by vrr

main

2026-08-28 12:20:33 +00:00

0s

fix(telemetry): attribute critical-path stall explicitly (#169)

#2 -Commit 7a8abcdda7 pushed by vrr

main

2026-08-28 02:29:19 +00:00

0s

feat(moe): hold back dense tensors larger than RAM under every mode

#1 -Commit 7b08da1dc7 pushed by vrr

v0.22.0-rc1

2026-08-28 02:42:19 +00:00

0s